Supplemental Information:
Direct recordings of grid-like neuronal activity in human spatial navigation
Joshua Jacobs1 , Christoph T. Weidemann2 , Jonathan F. Miller1 , Alec Solway3 , John Burke4 ,
Xue-Xin Wei4 , Nanthia Suthana5 , Michael Sperling6 , Ashwini D. Sharan7 , Itzhak Fried5,8∗ , & Michael J.
Kahana4∗
1
5
School of Biomedical Engineering, Science & Health Systems, Drexel University, Philadelphia, PA 19104
2
Department of Psychology, Swansea University, Singleton Park, Swansea, SA2 8PP, Wales, UK
3
Princeton Neuroscience Institute, Princeton University, NJ 08544
4
Department of Psychology, University of Pennsylvania, Philadelphia, PA 19104
Department of Neurosurgery, David Geffen School of Medicine and Semel Institute for Neuroscience and Human
Behavior, University of California, Los Angeles, CA 90095
6
Department of Neurology, Thomas Jefferson University, Philadelphia, PA 19107
7
Department of Neurosurgery, Thomas Jefferson University, Philadelphia, PA 19107
8
Functional Neurosurgery Unit, Tel-Aviv Medical Center and Sackler Faculty of Medicine, Tel-Aviv University,
Tel-Aviv 69978, Israel
∗ denotes equal contributions
Address correspondence to:
• Dr. Joshua Jacobs (Drexel University, [email protected])
• Dr. Michael J. Kahana (University of Pennsylvania, [email protected])
• Dr. Itzhak Fried (University of California, [email protected])
1
Nature Neuroscience: doi:10.1038/nn.3466
Patient #4 (H)
Count
0
0
0
1
Gridness score
p=0.011
Patient #6 (PHG)
0
Patient #7 (CC)
300
0
0
1
Gridness score
p=0.014
p=0.016
Patient #7 (CC)
Patient #10 (PHG)
300
0
Patient #7 (CC)
p=0.024
p=0.013
Patient #12 (H)
Count
0
0
p=0.034
0
1
Gridness score
p=0.003
0
Patient #11 (CC)
p=0.003
Patient #14 (EC)
Count
0
300
0
0
1
Gridness score
300
0
−1
2.0Hz
Patient #12 (H)
300
0
0
1
Gridness score
p=0.009
Count
300
0
1
Gridness score
p=0.026
0
1
Gridness score
p=0.014
0
1
Gridness score
p=0.028
Count
300
0
1
Gridness score
300
0
1
Gridness score
Count
300
0
0
1
Gridness score
300
0
1
Gridness score
Count
0
p=0.013
Count
p=0.005
Count
300
Patient #14 (EC)
0
0
1
Gridness score
Count
0
Patient #11 (CC)
300
0
1
Gridness score
Count
300
300
Count
Count
0
Patient #8 (H)
Patient #6 (EC)
Count
300
Patient #7 (CC)
p=0.018
0
1
Gridness score
p=0.049
Count
Patient #6 (EC)
300
Count
p=0.008
Count
300
Patient #14 (Cx)
300
0
1
Gridness score
p=0.024
Count
Patient #3 (EC)
0
1
Gridness score
0
0
1
Gridness score
Supplemental Figure S1: Additional cells that exhibited grid-like spatial firing. Each panel describes the
firing of a neuron that fulfilled our statistical criterion for exhibiting significant grid-like activity. Format for
the left and center panels follows Figure 2 in the main manuscript. Text in the upper-right corner indicates
the p value; label under the left panel indicates the plotted firing-rate range. Right panel illustrates the cell’s
true gridness score (dashed line) and the distribution of gridness scores from the shuffled data (bars).
2
Nature Neuroscience: doi:10.1038/nn.3466
A
First half
Second half
5 Hz
4.8 Hz
Correlation: r=0.25, p<0.001
B
1.1 Hz
1.2 Hz
Correlation: r=0.35, p<10-13
Supplemental Figure S2: Example grid-like neurons showing significant spatial stability. A. The activity
of a significant grid-like neuron from Patient 10’s entorhinal cortex in the first half (left) and the second
half (right) of the navigation session. This cell’s activity was significantly correlated betwen these two
intervals (p < 0.001). Top panels depict firing rate maps. Maps are plotted in a thresholded fashion, with
red coloring indicating regions with firing above a threshold and blue indicating below-threshold firing
(red/blue threshold labeled below plot). Bottom panels depict the cell’s spatial autocorrelation functions. B.
The activity of a significant grid-like neuron from patient 14’s entorhinal cortex, which exhibited significantly
correlated spatial firing between the two halves of the task (p < 0.001). Population stability analysis of grid-like
cells. We identified grid-like cells using only the first half of each session (to prevent bias) and computed
each cell’s firing-rate map separately for each half of the session. Then we used a Pearson correlation to
compute the pixel-by-pixel similarity between each cell’s two maps, assessing statistical significance using
a permutation procedure. We observed significant stability at p < 0.05 in 12% of the 49 first-half grid-like
cells (6 cells) and significant remapping in 0 cells. With a significance threshold of r > 0.2, 20% of cells were
stable (10) and 1 cell remapped. In both cases, the number of grid-like cells that were stable was reliably
more than the number that remapped (p’s< 0.02, binomial test).
3
Nature Neuroscience: doi:10.1038/nn.3466
B
Original data
C
Original data
Identify local peaks and compute
surrounding Voronoi polygons
Compute Fourier transform and
randomize spatial phase
Proportion grid cells
E:989:-@92+/:@,+F(**<+@2+GH
"#!&
"#!%
"#!
"#"D
"#"3
"#"&
"#"%
"
'()*+,)-)
7(.89:)**;+<=>??*(,
C8)-@)**;+C=>??*(,
Te
Re
Randomized data with
preserved spatial relations
al
mp
da
or
ta
Sp
all
ati
y
sh
all
ys
ys
uff
hu
led
hu
ffle
ffle
d(
d(
Vo
FF
ro
no
T)
ip
oly
go
ns
)
A
Sp
Rearranged
polygons
ati
all
Randomly move the firing associated with each
polygon to a new location and orientation
Rearranged
firing rate
...
Supplemental Figure S3: Spatial shuffling control analyses. These analyses tested whether our findings
of significant grid scores could be caused by place cells with multiple place fields that were not specifically
arranged in the 60◦ symmetric pattern of a true grid cell. A. Spatial-shuffling procedure based on a Fourier
transform. In this procedure we calculated two-dimensional discrete Fourier transform of each cell’s firing
rate map. Then we generated a series of shuffled surrogate datasets by randomly interchanging the phases
of each map’s Fourier components and then applying the inverse Fourier transform to calculate a phaseshuffled firing-rate map. This shuffled map has the locations of individual firing fields randomized without
significantly changing their size, magnitude, or number. We computed the gridness scores of the shuffled
firing maps and measured their significance using the methods from our main analyses. Plots indicate
examples of this procedure, with the top panel depicting an example cell’s true spatial firing and bottom
plots showing example randomized firing maps after shuffling. B. Spatial-shuffling procedure based on
global shuffling with preservation of local spatial structure. Following the methods of Krupic et al. (2012),
we computed each cell’s true firing map and then constructed a set of Voronoi polygons surrounding each
local firing peak. For each cell we translated and rotated the spiking associated with each polygon to a new
random location and orientation in the environment, and then computed the resulting gridness score. This
shuffling procedure served as a critical control because it allowed us to verify that the observed gridness
scores were truly related to global firing patterns rather than local firing variations, which were preserved
within each Voronoi polygon. C. Proportion of observed entorhinal cortex grid cells for true and spatially
shuffled datasets.
4
Nature Neuroscience: doi:10.1038/nn.3466
A
B
E
045_ses5
place P=0.01
J038_ses1 CSC19_1 LMH (regCode=2)
FR=1.4.CSC03_1
place P=0RAH (regCode=2)
place PFR=8.0.
p=0 clim=0−10
15
4
5
0
FR (Hz)
C
5
%place
placecells
cells
%
4
9
10
2
10
10
D
0
8
2
0.5
1
dir=1, place−X−dir=0
0
Firing rate (Hz)
5
2
5
5
1
0
0
N0W S E
Virtual Heading
dir=1, place−X−dir=0
*
7
6
5
place P p=0.01 clim=0−8.8
8
26
6
14
4
02
E
*
0
5
0
U406_ses2
RPHG (regCode=3)
FR=3.8.
place
P=0.61
E NW
S
_ses7 CSC15_1 RAH (regCode=2)
FR=10.3.CSC22_1
place P=0.9
dir=0,
place−X−dir=1
x direction
06_ses3 CSC22_1place
RPHG
(regCode=3)
FR=2.4.CSC19_2
place P=0RPC (regCode=5)
place PFR=0.7.
p=0 clim=0−8.9
4
04_ses3b
Heading
10place P=0.01
10
10
6
38
4
all place cells
region p=0.271958
place P p=0.01 clim=0−10
10
*
*
EC
H
25
2
PHG
A
CC
0
Region
direction cells
region p=0.000001
Cx
% dir cells
Supplemental Figure S4: Place cells. A. A place cell from patient 2’s left hippocampus. This cell’s activity
*
*
was significantly elevated during movement in one direction,
as detailed
in the inset. The color at each
20
location indicates the cell’s mean firing rate. B. A place cell from patient 5’s right hippocampus. C. A place
15 cell from patient 7’s right cingulate cortex. E.
cell from patient 9’s right parahippocampal gyrus. D. A place
*
*
Prevalence of significant (p < 0.05) place cells in each brain region; dashed line denotes the Type 1 error rate.
10
Black asterisks indicate regions where the number of observed
cells significantly
* exceeded chance levels.
*
5
0
5
Nature Neuroscience: doi:10.1038/nn.3466
EC
H
PHG
A
Region
CC
Cx
B
300
Count
A
300
0
−1
300
0
−1
0
1
300 Gridness score
0
−1
Count
0
−1
0
1
300 Gridness score
0
−1
Count
300
0
−1
0
1
300 Gridness score
0
−1
Count
Count
0
−1
0
1
300 Gridness score
0
−1
0
1
300 Gridness score
Count
0
1
Gridness score
0
−1
0
−1
0
1
Gridness score
300
0
Gridness score
Count
Count
0
−1
0
Gridness scor
Count
300
0
−1
0
1
300 Gridness score
0
Gridness scor
Count
Count
0
−1
0
1
300 Gridness score
300
0
Gridness scor
Count
Count
0
−1
0
1
Gridness score
300
0
Gridness score
Count
Count
Count
300
0
−1
0
Supplemental Figure S5: Subject movement in the virtual environment. Overhead maps of the environment with the subject’s movement path plotted as a gray line. Background coloration indicates the cell’s
firing rate at each location. A. The sub-panels of this figure are arranged to match the panels in Figure 2. B.
The sub-panels of this figure are arranged to match panels from Figure S1.
6
Nature Neuroscience: doi:10.1038/nn.3466
A
B
C
D
Supplemental Figure S6: Electrode localization. Figures depict the localization of a microelectrode bundle
in Patient 10’s right entorhinal cortex. A. Computed tomography (CT) scan after electrode implantation,
which shows the electrode positions. The CT image is then co-registered with pre-implantation structural
magnetic resonance image (MRI) scans to reveal electrode locations relative to anatomical landmarks. The
position of the microelectrode bundle in the left entorhinal cortex is shown by the green crosshairs. B.
Coronal MRI image. C. Sagittal MRI image. D. Axial MRI image.
7
Nature Neuroscience: doi:10.1038/nn.3466
Region
# cells
# grid-like
cells
# grid-like cells
with direction
sensitivity
# grid cells with
turn-sensitivity
# grid cells with
acceleration
sensitivity
# grid with angularacceleration
sensitivity
Entorhinal cortex
Hippocampus
Cingulate cortex
89
185
156
11
16
18
1
5∗
2
4∗
1
2
2
1
0
3
2
1
Supplemental Table S1: Grid-like cells that also exhibit direction, turning, acceleration, angularacceleration-related responses. Asterisks denote p < 0.01, Bonferroni corrected.
8
Nature Neuroscience: doi:10.1038/nn.3466
Region
Subject #
1
2
3
4
5
6
7
8
9
10
11
12
13
14
Total
EC
0 of 0
0 of 0
1 of 7
0 of 0
0 of 0
3 of 18
0 of 0
0 of 0
0 of 0
5 of 22
0 of 11
1 of 15
0 of 0
2 of 16
12 of 89
H
0 of 4
0 of 12
0 of 0
5 of 39
2 of 34
0 of 21
0 of 0
1 of 8
1 of 28
2 of 26
0 of 0
3 of 10
0 of 1
0 of 2
14 of 185
PHG
0 of 5
0 of 0
1 of 51
0 of 0
0 of 0
1 of 26
0 of 0
0 of 22
2 of 31
3 of 24
0 of 2
0 of 0
1 of 20
0 of 0
8 of 181
A
0 of 0
0 of 0
0 of 0
2 of 26
0 of 0
0 of 34
0 of 0
0 of 19
2 of 48
1 of 21
0 of 0
0 of 18
0 of 0
0 of 4
5 of 170
CC
0 of 0
0 of 0
0 of 0
0 of 0
0 of 0
0 of 0
13 of 113
0 of 0
0 of 0
0 of 0
4 of 30
1 of 10
0 of 0
0 of 3
18 of 156
Cx
0 of 0
0 of 0
0 of 0
0 of 0
0 of 0
0 of 0
0 of 0
0 of 5
0 of 23
0 of 0
0 of 5
1 of 20
3 of 43
1 of 16
5 of 112
Unique recording channels
0
0
2
3
1
3
11
1
4
9
3
6
3
3
49
Supplemental Table S2: Summary of grid-like cells across patients. Table indicates the total number of
cells and the number of grid-like cells for each patient, aggregated by brain region. The right-most column
indicates the number of grid-like cells observed on unique recording channels across sessions of the task.
9
Nature Neuroscience: doi:10.1038/nn.3466
Region
Mean
Firing
Rate
(Hz)
Firing Rate
5%–95%
range (Hz)
Entorhinal cortex
Hippocampus
Parahippocampal Gyrus
Amygdala
Cingulate cortex
Temporal cortex
2.8
3.5
2.6
1.7
3.2
3
0.01–9.9
0.03–14.8
0.06–8.9
0.09–5.5
0.1–14.3
0.1–11.9
Mean
spike
amplitude
(µV)
62.4
51.6
35
50.4
43.4
65.4
Background
noise
(µV)
Mean
false
positive
(FP) rate
FP
below
10%
FP
below
20%
Mean
false
negative
(FN) rate
FN
below
10%
FN
below
20%
14.4
9.6
7
9.4
9.3
16.4
9.6%
9.7%
11%
7.5%
6.9%
10%
71%
70%
65%
81%
83%
79%
87%
87%
75%
87%
89%
88%
14%
31%
17%
9.1%
9.9%
49%
73%
73%
77%
88%
85%
76%
84%
78%
80%
91%
87%
80%
Supplemental Table S3: Statistics on cell isolation quality for neurons from each region. Statistics
computed from each neuron’s waveform shape using methods described by Hill et al. (2011). False-positive
rate indicates the estimated percentage of spikes that were inappropriately designated as belonging to that
neuron (instead they came from noise or a neighboring cell). False-negative rate indicates the percentage
of spikes that were caused by a given neuron but were inappropriately labeled as members of neighboring
clusters or noise. Overall, the vast majority of cells had false-positive and false-negative confusion rates
below 10%, consistent with recordings from animals. We compared the spike waveforms of putative gridlike cells with those of other neurons and did not find any differences in their mean amplitude (48 µV
for grid cells vs. 46 µV for other cells; p > 0.95, t test) or in our false-positive or false-negative rates in
distinguishing their waveforms from neighboring cells (p’s> 0.5)
10
Nature Neuroscience: doi:10.1038/nn.3466