Science - USA (2022-01-07)

9 to 16 were delivered on the usual VR5 sched-
ule but were diluted by a factor of 4. This
caused a disruption in the established goal-
directed behavior without affecting consump-
tion behavior. Mice exhibited significantly
longer action period duration, interbehavioral
sequence duration, and latency to reward con-
sumption (Fig. 5, A and B) caused by increased
non-task-related behaviors both inside and
between behavioral sequences (fig. S14, A to C).
Eventually, all mice obtained and consumed
the maximum number of available outcomes.
We then tested the effect of the violation of
the outcome value on the neuronal represen-
tations of action and consumption at both the
single-neuron level and the population level
(326 neurons from four mice). First, whereas
action and consumption neurons detected
during the initial period showed significantly
reduced responses during the perturbed period
(Fig. 5, C and D), additional consumption and

action neurons emerged during the perturbed

period (fig. S15, C and D). Similarly, immedi-

ately after the start of outcome value violation

(i.e., after consumption of the first diluted re-

ward) new action- and consumption-associated

neuronal activity patterns emerged that re-

placed the action- and consumption-associated

neuronal activity patterns of the initial period

(Fig. 5, E and F, and fig. S15F). This effect was

also observed when considering only neuronal

activity patterns associated with the first or

the last action in an action period, indicating

that it was not a consequence of the increased

action period duration (fig. S15G, and see com-

parison with neuronal activity pattern stability

on day 5 in fig. S16).

During the violation of action-outcome con-

tingency paradigm, the initial eight rewards

were delivered contingent on the actions on

the usual VR5 schedule. Subsequently, out-

comes 9 to 20 were delivered noncontingently

on the animalÕs actions. Whereas at the be-

havioral level, this test disrupted goal-directed

behavior similarly to the violation of outcome

value (Fig. 5, G and H, and fig. S14, D to F), the

impact on BLA neuronal activity was different

(434 neurons from five mice). Action neurons

detected during the initial period showed a

reduction of their responses during the entire

perturbed period (Fig. 5I). However, unlike the

immediate effect of violating outcome value,

this change developed slowly after the start of

the perturbation and was not accompanied by

the emergence of additional action neurons

(fig. S15H). In accordance with this finding,

at the population level, the action-associated

neuronal activity patterns gradually lost their

correlation with the action-associated neuro-

nal activity patterns of the initial period, but

no new action-associated neuronal activity pat-

terns emerged during the perturbed period

(Fig.5Kandfig.S15J;alsoseetheresultsfor

the first and last action periods in fig. S15L). In

clear contrast to the impact of the violation of

outcome value, consumption-associated activ-

ity during the initial period after violation of

action-outcome contingency remained stable

along the entire perturbed period (Fig. 5, J to

L, and fig. S15K). As previously reported ( 17 ),

we were able to detect a population of con-

sumption neurons that gradually increased

their activity over the course of unexpected

reward deliveries (fig. S15I). However, the over-

all population activity pattern during consump-

tion remained highly correlated throughout

the entire session. Although changing outcome

value resulted in the immediate emergence

of new action and outcome representations,

action-outcome contingency violation resulted

in a gradual loss of the representation of the

action with no emergence of a new stable ac-

tion representation and no effect on reward

consumption representation.

Discussion

Our findings support a crucial role for the BLA

in the motivational control of goal-directed

behavior. At the time of goal-directed action,

BLA PNs integrated and encoded pursued

outcome identity, pursued outcome value, and

action-outcome contingency information. The

maintenance of such prospective, outcome-

specific, and updatable neuronal activity re-

flects a specific motivational state ( 18 , 19 ) that

differs from a primary motivational state

known to energize goal-directed actions in an

unspecific manner ( 1 , 20 ). Furthermore, this

state could be retrieved upon re-exposure to

the task, but its maintenance depended on

continued reinforcement. At the time of con-

sumption, BLA PNs represented current out-

come identity and value but not licking

behavior. This is consistent with the observa-

tions that BLA neurons encode distinct reward

features ( 21 ), including value ( 22 ), magnitude

Courtinet al.,Science 375 , eabg7277 (2022) 7 January 2022 5 of 13

C

0

2

4

6

*

GFP ArchT

action

ArchT

cons.

0

20

40

60

GFP ArchT

action

ArchT

cons.

** **

Duration (s)

0

20

40

D

Beh. sequence 16

9

8

1

Beh. epochs (sec)

Beh. sequence

1 OFF 89 ON 16

20

0

40

Action

20

0

40 ArchT

Action 1-2 UR lick 1-2 R lick 1-2

Beh. sequence

1 89 16

20

0

40

Consumption

20

0

40 ArchT

Laser-OFF

0 10 20

Time (s)

Laser-ON

Behavioral sequence

Action Consumption

OFF ON

OFF ON OFF ON

GFP GFP

*

A

0

40

80

Action #

GFP : actionAction Consumption

0

40

80

ArchT : actionAction Consumption

0

40

80

ArchT : consumptionAction Consumption

B

Actions per min

0

5

10

15

25

**

GFP ArchT

action

ArchT

cons.

**

Action period

**

GFP ArchT

action

ArchT

cons.

Action to lick

latency

Inter-behavioral

sequence

Duration (s) Duration (s)

Laser ON period

GFP ArchT

action

ArchT

cons.

Action # Action #

50 s 50 s 50 s

Laser-OFF Laser-ON

20

60

Action 1

Action 2

Unrewarded lick 2

Unrewarded lick 1

Rewarded lick 2

Transition

Rewarded lick 1

Idle time

Exploration

Fig. 4. BLA PNs activity is necessary for the maintenance of goal-directed actions.(A) Examples of
the effects of optogenetic manipulations on goal-directed behaviors on day 5 for GFP- and ArchT-expressing
mice. Black dots indicate individual actions; green dots, action period onset; vertical dashed lines, outcome
delivery. Colors indicate behavioral epochs. Shaded yellow areas indicate laser deliveries; double-headed
arrows, laser-ON periods (behavioral sequences 9 to 16). (B) Left to right, actions per minute, action period
duration, action to lick latency, and interbehavioral sequence duration during laser-OFF (behavioral sequences
1 to 8, white bars) and laser-ON periods, respectively (yellow bars; two-sided pairedttest for action per
minute and two-sided Wilcoxon signed-rank tests for the other variables comparing laser-OFF and laser-ON
periods, *P< 0.05; **P< 0.01). GFP,N= 20 mice in three cohorts × two outcomes (N= 12,N= 8 mice
with laser stimulation during action periods and consumption periods, respectively, with data pooled). ArchT
action,N= 12 mice in two cohorts × two outcomes (laser delivered during action periods); ArchT cons.,
N= 8 mice in two cohorts × two outcomes (laser delivered during consumption periods). Box-and-whisker
plots indicate median, interquartile, extreme data values, and outliers of the data distribution. (C) Proportions
of different behavioral epochs during laser-ON periods (color denote behavioral epochs as in (A); Black
dots denote significant increase in behavioral epoch duration;P< 0.01, two-sided Wilcoxon signed-rank tests
comparing laser-OFF and laser-ON periods; see raw values in fig. S11). (D) Mouse behavior during laser-
OFF and laser-ON periods while milk outcome was available (left). Colors denote behavioral epochs as in (A).
Also shown is the duration of the distinct behavioral epochs during the behavioral sequences for each
sequence along laser-OFF and laser-ON periods (right, two-sided Wilcoxon signed-rank test).

RESEARCH | RESEARCH ARTICLE