Science - USA (2022-01-07)

behavioral sequences (Fig. 3B). To quantify the
mapping between behavior and the popula-
tion activity vector, we averaged the pairwise
correlations between all population activity
vectors in action and consumption epochs. In-
trabehavioral epoch correlations were positive
for outcome-specific action and consumption
epochs and higher than interbehavioral epoch
correlations (action-consumption) and OFF task
correlations (Fig. 3C and fig. S6B). Throughout
the session, the population activity vectors
during action and consumption epochs re-
mained positively correlated and distinct from
OFF task population activity vectors (Fig. 3D
and fig. S6C), suggesting that specific neuronal
activity patterns robustly define distinct be-
havioral epochs in an outcome-specific man-
ner. We then used dimensionality reduction
[t-distributed stochastic neighbor embedding
(t-SNE); see the materials and methods] with
correlation as the distance measure to visually
demonstrate the distinction between the neu-
ronal activity patterns associated with the OFF
task phase and the two outcome-specific ON
task phases (fig. S6A). Mapping the population

activity vectors associated with the different

behavioral epochs to the embedded activity

space revealed discrete clusters corresponding

to distinct neuronal activity patterns occurring

at specific behavioral epochs (Fig. 3E and fig.

S6, B and C).

Defining action-associated neuronal activity

patterns as all population activity vectors cor-

related with the population response occur-

ring at the time of the first action of a

behavioral sequence (P< 0.01; see the mate-

rials and methods), action-associated neuro-

nal activity patterns were not restricted to

action epochs but had already started before

the first action of a behavioral sequence and

lasted until reward detection (Fig. 3F). Indeed,

the action-associated neuronal activity pat-

terns that were maintained for several seconds

throughout their respective action periods en-

compassedunrewardedlickandidletime

epochs but not exploration epochs (Fig. 3G and

fig. S6D). Therefore, BLA action-associated

neuronal activity patterns persistently repre-

sent information about the pursued outcome

from the initiation of the behavioral sequence

until reward detection, not just during action

performance. By tracking neurons across days,

we observed that BLA action-associated neu-

ronal activity patterns were reactivated upon

re-exposure to the task during both non-

reinforcedandreinforcedtests,buttheirac-

tivity was maintained only in the latter case

(fig. S7). By contrast, consumption-associated

neuronal activity patterns were restricted to

consumption epochs and were not detected

during unrewarded lick epochs, suggesting that

consumption-associated neuronal activity pat-

terns reflect reward detection rather than lick-

ing behavior (Fig. 3, F and G, and fig. S6D).

Using single-unit recordings and inferred

spikes extracted from deconvoluted signal using

CNMF-E,weconfirmedthattheseresultswere

not merely reflecting slow Ca2+dynamics (figs.

S8 and S9).

Specific motivational control of

goal-directed behavior

To test whether action- and consumption-

associated BLA activity is necessary for instru-

mental goal-directed behavior, we inhibited the

Courtinet al.,Science 375 , eabg7277 (2022) 7 January 2022 3 of 13

D

Z-score 0

1

2

OFF ON: milk

A

Time (s)

0 10 20

Time (s)

0 10 20

0 10 20 0 10 20

0 10 20 0 10 20

0 10 20 0 10 20

15

10

5

15

10

5

15

10

5

15

10

5

Behavioral sequence (sorted)

Behavioral sequence (sorted)

Consumption neuron

Transition neuron

Action neuron

Action 1 Consumption 1 B

Neuron

Probability

Z-score

Z-score

n = 137

n = 39

n = 170

-2 024 -2 0 24

-2 024 -2 0 24

Time from action (s) Time from consumption (s)

0

1

0

1

-2 024 -2 0 24

0

0.5

1

0

0.5

1

1.5

0.5

0

-0.5

Action

Lick

1

C 15%

4%

19%

62%

Action

Consumption

Transition

None modulated

59

60

18

71

24

75

15

3

21

Sucrose

Milk

1 z-s

1 z-s

1 z-s

**

Neuron # ** **

Neuron # Neuron #

OFF task ON task

5

10

15

20

25

5

10

15

20

25

5 10 15 20 25 5 10 15 20 25

Correlation

0

1

Neuron pairs (%)^0

50

100

a-ac-ca-c

No correlation

Negative correlation

Positive correlation

E

F

Action 2

Unrewarded lick 2Rewarded lick 2

Transition

Idle time

Exploration

0

0.5

1

F1 Score

AllMod.

Time (s) N Mod.

0 300 600 900

All

Modulated

None

Modulated

Real

OFF Action 1

period

Action 2

Rewarded lick 1 Rewarded lick 2

Cont

rol

**

50 s

**

**

**

Action 1

Unrewarded lick 1

Rewarded lick 1

Fig. 2. Distinct sets of BLA PNs encode specific behavioral epochs.
(A) Mouse behavior (top) and corresponding activity of three neurons (bottom)
during 20 behavioral sequences when sucrose outcome was available. The
behavioral sequences were aligned to action period onset (left) or to
consumption period onset (right) and sorted by action or consumption period
duration [colors at the top panels denote behavioral epochs as in (D)]. (B) Action
and lick probability (top, around action and consumption periods onsets),
individual neuron response of all task-modulated neurons (middle), and average
response by functional group (bottom;n= 137 action,n= 39 transition, and
n= 170 consumption neurons) aligned to action periods (left) and consumption
periods (right) onsets (N= 8 mice in two cohorts,n= 905 neurons recorded
on day 5). Shading indicates SEM. (C) Proportion (left) and outcome selectivity
(right) of the task-modulated neurons. (D) Average activity of 10 action neurons
(orange) and 16 consumption neurons (blue) during OFF and ON phases (top)

aligned to the mouse behavior (background colors). Also shown is the pairwise

correlation of activity for these neurons during the OFF and ON task phases

(bottom). (E) Percentage of correlated neuron pairs during ON task phases on

day 5 for action-action (a-a), consumption-consumption (c-c), and action-

consumption (a-c) neuron pairs (**P< 0.01). (F) Performance of a linear

decoder trained on the distinction between the five behavioral epochs depicted

using responses of all PNs (n= 80 ± 10, mean ± SEM neurons per mouse),

only the task-modulated neurons (Mod.,n= 33 ± 4 neurons per mouse), or a

random set of none-task-modulated neurons (N Mod.,n= 33 ± 4 neurons

randomly chosen to match the number of modulated neurons). Left, Example

mouse decoder results compared with the real behavioral description (top).

Right, MeanFscore for all mice (N= 8, two-sided pairedttest, **P< 0.01).

Box-and-whisker plots indicate median, interquartile, extreme data values, and

outliers of the data distribution.

RESEARCH | RESEARCH ARTICLE