portions of the tree were more strongly cor-
related than others. Similar results were ob-
tained when event clustering was based on the
spatial extent of calcium signals instead of
their averaged amplitude (fig. S5).
To further examine the R/L hemi-tree
segregation, we used two additional analysis
methods: First, we shuffled the“within”and
“between”tagging of ROI Pearson correla-
tion pairs while maintaining their pairwise
correlation coefficient values. We found the
experimental values of the difference between
the mean Pearson correlations of within com-
pared to between hemi-trees to be significantly
higher compared to the shuffled distribution
for all cluster events (Fig. 2M and fig. S6, A and
C; Z-score >3.6). Second, we calculated the
proportion of the variance explained (R^2 ) ofthe calcium activity of each ROI in one hemi-
tree by the activity of all ROIs in the contra-
lateral hemi-tree using a linear regression
model. The results further indicated compart-
mentalization of activity in the hemi-trees in
clusters 1 to 3 (fig. S7A).
To facilitate the comparison between type
1 and type 2 PTNs, we also divided the events of
type 2 PTNs into four clusters. The distributions272 15 APRIL 2022•VOL 376 ISSUE 6590 science.orgSCIENCE
Fig. 5. Modeling the contribution of
the behavioral predictors to cal-
cium activity of type 1 and type
2 PTNs using a generalized linear
model (GLM).(A) Goodness of fit
(R^2 ) of GLM full model for each ROI
from a type 1 PTN during a hand
reach session. (B) Mean relative
contribution of modeled ROIs (R^2
15%) for each of the behavioral
predictors for right and left hemi-
trees. Black asterisks on both graphs
indicate values computed from
somatic recordings of same neuron.
(C) Structural distance matrix. ROIs
withR^2 < 0.15 were excluded.
(D) Pairwise Pearson correlation
between the GLM relative contribution
vectors for all included ROI pairs,
arranged by the tree structure.
(E) Pearson correlation values shown
in (D) as a function of shortest path
distance fitted with a linear regression
model. (FtoJ) As in (A) to (E) during
running on a treadmill. Same neuron
in (A) to (J). (KtoT) As in (A) to (J)
for an example type 2 PTN. (U) (Left)
Frequency of ROI’sR^2 values of GLM
full model for type 1 PTNs (10
animals, 14 neurons, 31 sessions).
(Right) As in left panel for type 2
PTNs (9 animals, 10 neurons, 31
sessions). (VtoX) Box plots of the
following parameters:R^2 linear
regression model that predicted the
Pearson correlations of GLM relative
contribution vector by dendritic dis-
tance (V), Mantel statistics comparing
the structural distance matrix and the
behavioral-correlation matrix (W),
Pearson correlation between the
soma’s behavioral relative contribu-
tions and those of the tuft, for
type 1 and type 2 PTNs (seven and
four neurons, respectively) (X). ***p<
0.001; blue asterisks, mean value.
Wilcoxon rank test.
GLM type 1 PTNsMantel = 0.596 (p< 0.001)Components
Behavioral correlation matrixROIs122011 22ROIsR ² =0.355, Slope=-0.5060(^2004006008001000)
0.2
0.6
1
-0.2
Correlation
Distance (μm)
E
A
Relative contribution (%)
ROIs
Full model
Explained variance^0
0.4
0.6
0.8
0.2
(^122)
B
C Structural distance matrix
ROIs
ROIs
1
22
1 22
900
(μm)
0
D
Hand reach
At mouthBack toFace map
Grab
Kinematics
Lift
Reward
Supination
Tone
0
10
20
30
Left hemi-tree
0
10
20
30
Right hemi-tree
40
Left
Right
Relative contribution (%)
Right hemi-tree
OnsetOffsetSpeed²
Speed³Accel²Accel³Speed
AccelWalkRest
Pos-accelNeg-accelLocation
0
20
40
0
20
40 Left hemi-tree
Components
ROIs
Explained variance
Full model
0
0.2
0.3
0.4
0.1
(^125)
R^2 =0.194, Slope=-0.537
Correlation
0.2
0.6
1
-0.2
0
(^2004006008001000)
Mantel = 0.554 (p< 0.001) Distance (μm)
1
15 1
1 15 -0.4
ROIs
ROIs
Structural distance matrix
900
(μm)
0
1
15
1 15
ROIs
ROIs
FGTreadmill
H
IJ
Behavioral correlation matrix
GLM type 2 PTNs
Mantel = -0.009 (p = 0.477)
R ² =0, Slope=0.007
(^050150250350450)
0.2
0.4
0.6
0.8
1
Correlation
Distance (μm)
Components
ROIs
(^1) 0.1
23 1
ROIs
1 23
Structural distance matrix
400
(μm)
0
1
23
ROIs
ROIs
1 23
Full model
ROIs
Explained variance (^127)
0.05
0.15
0.25
Relative contribution (%)
Right hemi-tree
Tone
Lift
Grab
SupinationAt mouth
Back toReward
0
10
20
30
40
Face mapKinematics
Left hemi-tree
0
10
20
30
40
KL
M
NOBehavioral correlation matrix
Explained varianceROIs
0
0.2
0.4
0.6
(^121)
Full model
R^2 =0.007, Slope=-0.059
100 200 300 400
0
0.2
0.4
0.6
0.8
1
0
Distance (μm)
Correlation
Right hemi-tree
0
10
30
20
Left hemi-tree
0
10
30
20
OnsetOffsetSpeed²Speed³Accel²Accel
³
SpeedAccelWalkRest
Pos-accelNeg-accelLocation
Components
Relative contribution (%)
Mantel = 0.085 (p = 0.092)
1 0.4
21 1
1 21
ROIs
ROIs
ROIs
Structural distance matrix
ROIs
1
21
350
(μm)
0
1 21
PQ
R
ST
Behavioral correlation matrix
Behavioral correlation vs.
dendritic distance
R²
Type 1 Type 2
1
0.4
0.6
0.8
0.2
0
V Behavioral correlation matrix vs
distance matrix
1
0.4
0.6
0.8
0.2
0
Type 1 Type 2
Mantel
Type 1 PTNs W
Explained variance (GLM)
0.05
0.15
0.25
0.35
0 0.20.40.60.8 1
Type 2 PTNs
0.1
0.2
(^00) 0.20.40.60.8 1
0.4
0.3
Treadmill
Hand reach
Explained variance (GLM)
Frequenc Frequency
y
U GLM vector Pearson correlation
tuft vs soma
R
Type 1 Type 2
1
0.4
0.6
0.8
0.2
0
- X
Treadmill
Hand reach
RESEARCH | RESEARCH ARTICLE