neuronal representations were outcome spe-
cific, behaviors of the two ON task phases were
similar, suggesting comparable outcome values.
Action and consumption neurons were oppo-
sitely modulated during ON task phases: Neu-
rons activated during action were silent during
consumption and vice versa (Fig. 2B). Accord-
ingly, pairwise correlation and coactivation
analysis of neuronal activity of action and/or
consumption neurons indicated that coinci-
dent activity within each functional subset
increased and coincident activity across func-
tional subsets decreased during ON task phases
(Fig. 2D). These patterns of coincidence were
limited to the ON task phase, suggesting task-
dependent functional interactions (fig. S4, E
and F). We confirmed the results obtained with
Ca2+imaging using single-unit recordings (n=
78 neurons from four mice). Similar to the im-
aging results, 28% of all recorded neurons
exhibited significant task-related modulation
(P=0.10andc^2 = 2.7, Friedman test). Corre-
lation and co-firing analysis of pairwise neuro-
nal activity between action and/or consumption
neurons confirmed task-dependent functional
interactions (fig. S5).
Because the activity of action and consump-
tion neurons was outcome specific, we tested
the performance of a linear decoder trained
with the imaging dataset on the five-way dis-
tinction among action periods, consumption
periods (milk or sucrose), and periods of non-
task-related behavior during the OFF task
phase. Decoder performance was high when
trained with activity traces of all PNs (F1 score =
87.8 ± 3.2%) and dropped to chance levels
when trained on shuffled neuronal data (19.7 ±
0.1%,P< 0.01, pairedttest; Fig. 2F and fig.
S4G). When subsets of PNs were selected, per-
formance was higher for training with the ac-
tivity of all task-modulated neurons (69.7 ±
6.2%) than for training with random selec-tions of non-task-modulated neurons (57.3 ±
5.1%; see the materials and methods).BLA activity patterns tile the entire
action-consumption sequences
To investigate population-level representation-
al principles and contrast them with our ob-
servations at the level of single neurons, we
obtained population activity vectors by bin-
ning single neuron activity in 200-ms bins.
Population activity vectors time locked to the
beginning of action and consumption epochs
showed regularity and specificity (pairwise
Pearson’s correlations action-action,r=0.24±
0.02; consumption-consumption,r=0.52±
0.02; action-consumption,r=–0.004 ± 0.01
for this example mouse; Fig. 3A). According-
ly, pairwise correlations between population
activity vectors along the entire behavioral
session mirrored both the different task phases
and the behavioral epochs in action-outcomeCourtinet al.,Science 375 , eabg7277 (2022) 7 January 2022 2 of 13
OFF ON: milk OFF ON: sucrose OFFDF Behavioral sequence
Action
periodConsumption periodTraining daysA54321
Action 1 Lickports Action 2
CR
VR3
VR5 Actions per min
05101520Training days123 4 5 Day 8Task
phaseAction #040608020Time (s)0 100 200 300 400 500 600 700 800 9000 10 2010515
Time (s)Beh. sequence (sorted)E# of action per action period0 5 10020
1030Action period duration (s)Inter-behavioral sequence5 1001020HTime (s)600 650 700 750 800Neuron #10155phaseTa s k OFF ON: sucrose040
20
Behavioralepochs (s)
5 10
Behavioral sequenceUR lick 1-2Action 1-2
R lick 1-2Instrumental training Shift in
hunger state nsCAction (%)
040608010020Action 1Action 2Extinction
***Dev.
Non dev.Action (%)
040608010020Devaluation
**Deg.
Non deg.Action (%)
040608010020Degradation20%Day 518%
11%
3%30%5%5%
3%
5%Day 1
4%2%14%13%10% 32%22%1%2%ON task phases051002
13Actionrate (Hz)BGDuration (s)Behavioral sequence5 10024Action to
lick latency (s) 1515 152 z-sAction 1Action 2
Unrewarded lick 2Unrewarded lick 1Rewarded lick 2TransitionRewarded lick 1Idle time
ExplorationFig. 1. Ca2+imaging of BLA PNs during the learned, self-paced,
instrumental goal-directed task.(A) Schematic of the behavioral context.
Action availability is controlled and the required number of actions per
outcome increases over training from CR to VR5. (B) Number of actions
per minute over training (N= 16 mice in two cohorts) and during the free
choice rewarded session in sated mice (N= 8 mice in two cohorts). (C) Action
choice during the extinction test, satiety-induced outcome devaluation test,
and action-outcome contingency degradation test (all tests were non-reinforced
from two mice cohorts;N= 16,N= 13, andN= 10 mice,P= 0.72,P< 0.001,
andP< 0.01, respectively; two-sided pairedttests comparing action choice).
Box-and-whisker plots indicate median, interquartile, extreme data values, and
outliers of the data distribution. (D) Task phases and behavioral epochs labeling
during day 5 for one mouse. Black dots indicate individual actions; red and
green dots, action periods onset (for milk and sucrose, respectively); vertical
dashed lines, outcome delivery. Inset, initial behavioral sequences. Colors
indicate the different behavioral epochs. (E) The proportion of time mice
engaged in the different behavioral epochs during ON task phases on training
days 1 and 5 (N= 16 mice; black or white dots denote significant increase
or decrease in behavioral epochs duration, respectively;P< 0.01, two-sided
Wilcoxon signed-rank tests comparing day 1 and 5 epoch durations).
(F) Behavioral sequences aligned to action period onset (0 s) and sorted
based on action period duration (left) while milk outcome was available.
Duration of action periods (r= 0.21;P< 0.001) and action rate within action
period (r=Ð0.07;P= 0.1) across action periods (top right). Mean duration
of the distinct behavioral epochs (bottom right,N=16mice,day5).(G) Action
to lick latency and duration of the interbehavioral sequence intervals across
behavioral sequences (N= 16 mice, day 5). Shading indicates SEM. (H) Ca2+
traces from 18 BLA CaMK2-expressing neurons simultaneously imaged during
4 min of day 5 [same mouse as in (D)]. Scale bar, 1z-score. Colors at the top
panels denote the different behaviors as in (D).RESEARCH | RESEARCH ARTICLE